Methodology
From data to chart
- Collection. The pipeline downloads data from official sources and stores each raw file with its URL, collection date, license, source version and sha256 hash.
- Normalization. Everything becomes one format: geography, year, indicator, value, confidence interval when available, source, version, license and a series-break flag (ADR 0002).
- Validation. Schema, per-indicator bounds, missing years and atypical jumps. Real jumps, such as hyperinflation and the pandemic, become warnings for review, never corrections.
- Publication. The site is generated only from files versioned in the repository. CI checks that regenerating everything from raw data reproduces exactly the published data.
Traceable numbers
No number is typed by hand in the texts. Numbers come from repository data, from peer-reviewed papers (with DOI and location in the paper) or from a written justification. Every number in the text links to the "where each number comes from" table. DOIs are checked on Crossref (ADR 0007).
Governments and causality
Government bands on charts and the per-period change table are descriptive. A year is attributed to the government in office on July 1. Policy effects lag, external factors matter, and Congress, the states and the central bank also decide. That is why Brazil is always shown alongside neighboring countries, and causal claims appear only in analyses with a causal design (ADR 0003). To see what each type of study can show, read how to read studies.
Series breaks and uncertainty
When a survey's methodology changes, as when PNAD was replaced by the Continuous PNAD, series are not spliced: they are shown separately, with the break marked on the chart. Confidence intervals appear as light bands when the source publishes uncertainty.
Neutrality
Every analysis includes evidence for and against its hypothesis. Every policy brief includes a null, negative or mixed result, or documents where contrary evidence was sought.
Licenses
Each observation carries its source's license (ADR 0006). All methodological decisions are in docs/adr. The data changelog is in data/CHANGELOG.md.