Why these numbers can be checked
We do not ask you to trust us. Each claim below points to where you can check it, and the weaknesses are listed as carefully as the safeguards.
A HumanToll figure cannot be shown without its range, its scenario, the calculation it came from and the versions used. It is published only from a calculation that passed a test on past years. Its breakdown must add up to it exactly. And anyone can report an error and read the answer in public.
What each figure stands on
No figure without its range
The one piece of software that prints a figure refuses to work unless it is also given the calculation, the scenario, the low and high ends of the range, the confidence grade, and the versions of the method and the data.
A test on past years before anything is published
The model is run for every year from 2010, on two different tables of how industries buy from each other and under five scenarios, and judged by twelve checks. A calculation can be published only if it passed them all and the result was recorded.
A breakdown that must add up
Every figure is broken down by material, country, war and kind of harm, and every breakdown must add up to the figure. If one does not, the page refuses to show it rather than hiding the difference.
Two methods, never added together
Working from what a product is made of, and working from the industries that make it, answer two different questions. Both answers are shown; they are never added into one number.
A dated, versioned method
The method is published before any figure, and each finished version stays online unchanged, so a citation always points to the method that produced the number.
Licensed, credited sources
Every source is listed with its licence and the exact version used. Where a publisher gave permission in writing, the letter is kept on file.
A public corrections log
Anybody can report an error. Reports are answered in public, the rejected ones too, and an accepted correction appears on the page it corrects.
As of this page
- Published estimates: yes.
- The latest test on past years judged 140 runs: 140 publishable (a run is one calculation).
- Corrections: 0 reported, 0 answered, 0 waiting to be read.
Where the model is weakest
- Harm that cannot be linked to a product is left out, so every figure is a floor, not a total.
- Only making a product is counted. For a car, the harm from using it is probably the larger part.
- Deaths caused indirectly by conflict — disease, hunger, displacement — are not counted.
- Injury figures depend on who reports them: when a country stops reporting, its injuries disappear from the figures.
- Measured per euro spent, most industries come out too close to tell apart, so ranking them that way means little.
- Several of the five scenarios still give the same figures, because the assumptions that set them apart have little to work on yet.