Ourus · a decision story
The budget fellby half.The need did not.
US official development assistance fell from about sixty-three billion dollars to under twenty-nine billion in a single year. The UN's 2025 appeal took in twelve billion — the lowest in a decade. Twenty-five million fewer people were reached than the year before.
This is not our framing
Triage.
239M
people in need — Global Humanitarian Overview 2026
135M
targeted, at $33 billion
87M
"hyper-prioritized," at $23 billion
27.8%
how much of the 2025 appeal was funded
The current method
A map, a spreadsheet,and a room.
Site selection is a workshop. Local knowledge, a printed map, available land, and whoever argues hardest. It is not negligence — it is the absence of a tool. The people doing it are the best-informed people in the country, working without the one thing that would settle the argument.
What is missing
Three decisions,one budget.
Where the sites go, before the season starts. Which modality reaches which caseload, while it is happening. What moves, after a slope fails. Three products, sold in that order.
The case
Cox's Bazar.The monsoon before last.
One million people across thirty-three camps, on erodible hills, in the most densely populated refugee settlement on earth. A hundred and fifty-two thousand new arrivals since late 2023, against a response plan written for fifty thousand. In one week of July 2026: two hundred and eighty-six weather incidents, ninety-five landslides, twenty-six thousand people affected.
Before the season
The standards arethe constraints.
Fifteen litres per person per day. No household more than five hundred metres from a water point. No more than twenty people per toilet. Three and a half square metres of covered space each. These are published, numeric, and binding. That makes siting a coverage problem with a budget — not a workshop.
The deliverable
A site plan she canhand to a donor.
Nine positions, four asset classes, a coverage number per dollar, and every Sphere clause it satisfies printed against it. Defensible line by line in a funding conversation — which, this year, is the conversation.
Why the simulation half matters
People move towardwhat you built.
Standard planning treats beneficiary locations as fixed input, solves against them, and is wrong by the second month. Put in a water point and the settlement pattern shifts toward it. The plan changes the thing the plan was solved against.
What we model instead
Three behaviours,not three thousand draws.
Every one of these is a household making a rational choice. That is precisely why a fixed distribution cannot represent them, and why the plan has to be tested against them before the rain.
Before the rain, not after
The plan fails in a model,not in a monsoon.
Run the three behaviours forward nine weeks against the plan we just solved, and the gap has a number and a location: one thousand three hundred and forty households beyond five hundred metres. Two hundred and eighty-six incidents in six days is not a surprise. It is a season.
After a loss
The slope goes at 3 a.m.The re-plan cannot take a month.
Drop the lost water points. Re-weight what survives — because a blocked path just made one surviving tapstand serve nine hundred people against a standard of two hundred and fifty. Re-solve in seconds. Nobody re-solves for week ten.
What we are asking for
Not a demo.A dataset.
We are two engineers in Toronto. Neither of us has run a camp. What we have is a set-cover solver that placed charging infrastructure for the Ontario Ministry of Transportation, and a simulation that models a population as something that moves. Give us one camp, one season of data, and the plan you already wrote — and let us show you where we are wrong.
Behaviour set
The arrival152,000 since late 2023
Arrives against a plan written for fifty thousand. Settles in the gap between two blocks, because that is where there is room.
The shortest pathevery household, every day
Walks to the nearest water, not the assigned one. Queues form where the model did not put them, and the fifteen-minute standard fails at a tapstand costed for two hundred and fifty.
The slope4,307 shelters
When the flat land is taken, the next household builds on the gradient. Your plan's own success moves people onto the ridge that fails in July.
One settlement, one season, one budget
Within Sphere standard—
Longest walk to water—
People reached, one week—
Funding figures: OECD preliminary 2025 ODA; UN OCHA Global Humanitarian Overview 2026 (Dec 2025). Cox's Bazar figures: UNHCR, IOM DTM, and Human Rights Watch (July 2026) — 286 weather incidents and 95 landslides, 4–9 July 2026. Standards: Sphere Handbook 2018, 4th edition, and UNHCR Emergency Handbook. Modality efficiency: CALP Network (2026) and ODI four-country study. Settlement-drift evidence: Migration Policy Centre. Site plan is a stylised reconstruction. All solver output is illustrative.
OURUS
The same money. More people.
Placed by math. Stress-tested against a population that moves.
OURUS.AI·CONTACT@OURUS.AI
An illustrative reconstruction. Ourus has not worked in Cox's Bazar. Settlement geometry, costs, coverage figures and re-solve results are model output on synthetic data, not operational data from any organisation. Context figures — population, incident counts, funding levels and Sphere standards — are from published sources and cited on screen.