Conflicts
Attributing a death to an economic sector is the step this project will be attacked for, so it is scored in the open: nine criteria, a published rubric, a citation per score, and five scenarios that filter the same scored dataset in five different ways.
Twelve conflicts have been scored so far, chosen to span the argument rather than to be easy — a war financed by artisanal mining, one financed by diamonds on one side and offshore oil on the other, a genocide with almost no economic content, and a petro-state whose war has nothing to do with its oil. The ledger holds over a thousand conflicts. That gap is on this page because a sample presented as a survey is the easiest way to mislead with true numbers.
Scored conflicts
Ordered by the number of deaths the publisher recorded, not by anything this model computed — an index ordered by a model output would imply a ranking the model has not earned. A conflict with two periods has been scored twice, because its economy changed.
| Conflict | Type | Deaths recorded | Scored periods | Sectors | Scores |
|---|
Death and event counts: Uppsala Conflict Data Program, Georeferenced Event Dataset v26.1, CC BY 4.0.
How much of the ledger this covers
0 conflicts are scored. The ledger holds 1,581.
The twelve were chosen to span the argument, not to cover the ledger: a resource war, an insurgency financed by artisanal mining, a state collapse, a proxy war, an interstate war, two conflicts whose economics are genuinely thin, and one whose economy is a criminal market no classification contains. Between them they exercise every part of the scoring machinery, including the parts that are supposed to refuse.
Scoring nine criteria across several sectors against evidence is research, not data entry, and it is the slowest thing in this project. Until the coverage is much larger, no total across conflicts is published — a sum over twelve conflicts presented as a global figure would be wrong by roughly the amount it left out.
The five scenarios, and the three flags that separate them
One scored dataset, five filters. Every difference between the scenarios is in this table; there is no scenario-specific scoring anywhere.
| Scenario | Evidence required | Confidence floor | Sectors admitted | Other propagates |
|---|---|---|---|---|
| Economic-embedded S-MAX |
no | LOW | every named sector | yes |
| Evidence-weighted S-EVID |
yes | LOW | every named sector | no |
| Resource-only S-RES |
yes | LOW | ENERGY, MINING, AGRICULTURE, LAND, WATER, FORESTRY | no |
| Arms and resources S-ARMS |
yes | LOW | ENERGY, MINING, AGRICULTURE, LAND, WATER, FORESTRY, ARMS, MILITARY_PROCUREMENT | no |
| Conservative S-CONS |
yes | HIGH | every named sector | no |
No scenario renormalises. A share a scenario declines to route moves to Other and stays in the total, so an evidenced share keeps the same magnitude in all five and the gap between them is a published number rather than a rescaling nobody can see.
Questions
Why is “Other” so large for some conflicts?
Because for some conflicts it should be. Other holds the economic activity no named sector accounts for, and it is scored like any sector rather than being a remainder. A separatist war over citizenship, or a genocide organised administratively and carried out where people lived, has little productive geography — and a model that could not report that would be forced to invent an economic story for the cases that most obviously do not have one. Every conflict whose Other exceeds 30% carries a written review saying whether it is under-scored or genuinely outside the named sectors.
Does a high share mean the sector caused the war?
No. It means the conflict’s human toll is carried by that sector for the purpose of tracing harm through a supply chain — which is a claim about accounting, not about motive. The default scenario is explicit that it “quantifies the economic activity associated with conflict rather than the motivation of anyone fighting it”.
Who did the scoring?
It is seeded, versioned and auditable: every score carries the source it was read from, when it was entered and by whom, and any later change is written to an append-only audit table that takes no update and no delete. If you think a score is wrong, the rubric anchors and the citation are both on the conflict’s page, which is the point of publishing them.
Why are only conflicts scored, and not workplaces or disease?
Occupational harm is already reported by economic sector, so it needs a concordance rather than a judgement. Deprivation and exposure harm needs the same nine criteria applied to a different kind of subject, and it comes later in the build. The machinery on this page is what all four channels will use.