Method
The value of a reference is that you can trust it when you look something up. These are the rules that make that true, and the ones that limit what the site will tell you.
Every figure comes from a national statistical agency or an international body that compiles them. No opinion polling, no composite indices, no third-party summaries. Each page names the dataset and the exact indicator codes, and links to the query that produced the numbers, so any figure can be checked at its source.
Sources currently in use: World Bank World Development Indicators.
Figures are derived from the published series at build time. Nothing numeric is written by hand, including the figures in the chart captions. This is not fussiness: an early hand-read of the life-expectancy data put the widest sex gap at 1970 and 7.6 years, when the answer across all 65 observations is 1968 and 8.1 years. Computed pages cannot make that mistake; written ones do.
Where a year reports one sex and not the other, the year is dropped rather than charted — a half-populated year would otherwise plot as a gap of zero.
Every comparison carries both. They are required fields in the schema, and the build rejects a record that has fewer than two entries under what it does not say. The constraint is mechanical on purpose: a caveat that is optional gets skipped on exactly the pages that most need it.
Most statistics are misread in a predictable way, and that predictable misreading is what the section is for. Life expectancy is not a forecast. An unemployment rate is a share of the labour force, not of the population. A gross enrolment ratio can exceed 100% without being wrong.
The site shows what is counted. It does not say why a gap exists, whether it should, or what anyone ought to do about it. That is a deliberate limit, not modesty: the moment a reference starts explaining its own numbers it becomes a position, and a position is not something you can look things up in.
Causal questions are the interesting ones, and they are answered in the research literature, not on a page like this.
The comparisons follow the categories the source data uses. Vital statistics and labour surveys report male and female, so that is what these pages report. Where a dataset carries a third category — the Census Annual Business Survey counts equally male- and female-owned firms as their own class — it is shown rather than folded into one side or dropped.
Agencies revise. Each page records the date the source itself was last updated and the date the data was retrieved, which are different things and both matter. The most recent year of any series is usually an estimate and should be read as one.