Methodology

We harvest from multiple source families: official government statistics (FBI UCR, Eurostat), global conflict databases (ACLED, UCDP, GDELT), attitudinal surveys (Pew, World Values Survey), NGO trackers (SPLC, Human Rights Watch), and peer-reviewed scholarship.

No single source dominates any page. When official data and community reports diverge -as they often do with hate crime statistics -we show both and explain the gap.

Every claim, chart, and timeline entry carries a confidence level derived from:

• •Source diversity -how many independent source families corroborate the claim
• •Coverage -geographic and temporal completeness of the underlying data
• •Recency -when the source was last updated
• •Methodological rigor -peer review, sample size, reporting methodology

“Insufficient evidence” is a first-class state. We never force a score when data is inadequate.

We maintain a strict separation between inbound hostility (directed at a group) and outbound hostility indicators (hostility expressed by actors associated with a group or context). Attitude, violence, policy, and rhetoric are kept as separate dimensions before being combined into any composite view.

If official crime statistics show low hate crime rates but survey data shows high experienced discrimination, we present both perspectives with explanatory context about underreporting, definitional differences, and institutional gaps. We never flatten complexity to produce a simpler narrative.

Hate crime statistics systematically undercount actual incidents. FBI UCR data relies on voluntary agency participation (covering ~85% of the population). Victims often do not report. Legal definitions vary by jurisdiction. We surface these caveats alongside every data point drawn from official statistics.

Content follows a six-stage pipeline: source ingestion, normalization, claim extraction, confidence scoring, human review and narrative synthesis, and publication with changelog. Every published page carries a last-updated date and version history.

• ×Predict what a specific person believes
• ×Rank countries or groups as “most hateful” without context
• ×Use gamified scoring on painful material
• ×Present harm without linking to bridges and reform stories