Read the math. Then cite it.

The public method explains the sources, normalization logic, coverage limits, refresh cadence, and errata policy without exposing implementation details.

Safety Atlas is an independent research index. We do not sell access to public data — that is free. We sell packaged analysis, documentation, and licensed deliverables that make the data usable for business and editorial decisions. Deeper audit packs are scoped privately.

Version 1.0 · Issued 13 May 2026 · Source: FBI NIBRS / UCR + Census ACS. No real-time feed, predictive policing model, or individual-level data.

Per offense, on a 0–100 scale.

Each offense is normalized against the national baseline for the same offense and year, then converted to a 0–100 score. Lower reported crime rates produce higher safety scores.

The public product does not publish internal formulas or coefficients. Buyer-facing files disclose source, coverage, year, and confidence notes so the result can be cited without exposing the scoring implementation.

Three categories, weighted.

Offenses are grouped into personal safety, property security, and community order. The composite gives greater importance to person-directed offenses while still retaining property and public-order signals.

Detailed weighting and sensitivity tests are available only in scoped audit deliverables, not on the public site.

A small-town relief curve.

Raw per-capita crime rates can overstate small-town risk. Safety Atlas applies a population adjustment so very small jurisdictions are not over-penalized by one-off incidents.

The adjustment is monotonic and disclosed at a principle level; the exact curve remains part of the licensed methodology.

When numbers move, who pays, who knows.

Refresh: annual full refresh in Q2 (with the FBI NIBRS annual release), quarterly delta releases for state-and-larger aggregations. This is not a real-time feed. If you need real-time, this is not the product — see Cherre or HazardHub.

Errata: any factual error in published numbers triggers a public changelog entry and a proactive refund to every buyer affected in the prior 30 days. We do not silently correct. We do not retroactively rewrite published reports.

Refunds: 7-day, no-questions, one-click via reply to the receipt email. After 7 days, refunds are case-by-case but methodology disputes are always refunded if we cannot defend the math in writing.

Where the data is thin, we say so.

FBI NIBRS adoption is not 100% uniform. A small number of agencies report inconsistently or have stopped reporting. Coverage is published per state in the State Scorecard and per city in the city report. We never extrapolate from missing data. We never impute. If a city stopped reporting in 2023, we score it through 2022 and mark it discontinued.