American Context

ATLAS · QUESTIONS

Golden civic questions.

The civic questions that recur — across spending, disasters, legislation, demographics, health, and crime — get permanent pages with the structured agent plan, primary citations, and reproducible reasoning trail.

Question pages are eval contracts that became prose

Every public question page is built from the same retrieval plan and tool sequence the agent uses for live queries. The page is the answer; the page is also the evaluation. If the underlying corpus changes, the page rematerializes and the eval reruns.

Questions become “golden” only after they clear deterministic fact checks, claim audits, and source-citation gates. Pages stay in preview until reviewer approval under the Atlas G6 workflow.

What every question page contains

  • The question

    Plain-language phrasing as a person would actually ask it.

  • Agent plan

    The retrieval and tool sequence the agent executes, made visible.

  • Expected sources

    Primary government data the answer must cite to be valid.

  • Structured answer

    Numbered, citable facts derived from the agent's tool outputs.

  • Limitations

    Source gaps, freshness lags, and inference boundaries — visible by default.

  • Eval contract

    Deterministic checks each generated answer must pass before publication.

Sample golden questions

Each represents a civic question that recurs often enough to warrant a permanent, citable page. These mirror the agent's evaluation corpus.

  • Federal spending

    ›How much has the federal government obligated under FEMA Public Assistance in Florida since 2020, and which counties received the most?

    Combines program authority, fiscal obligations, and county-level award lineage in one answer.

  • Disaster equity

    ›Which West Virginia counties received the least disaster assistance per capita after the 2016 federal disaster declarations?

    Surfaces equity gaps between recovery dollars and demographic exposure.

  • Legislation impact

    ›Which Infrastructure Investment and Jobs Act programs have obligated the most money in the first three years, and to which states?

    Maps statutory authority through Treasury obligations to recipient geography.

  • Climate and disaster

    ›Which Florida counties affected by Hurricane Ian have active FEMA Public Assistance projects, and what is each county's obligation total?

    Joins disaster declarations, FEMA award rolls, and county-level fiscal totals.

  • Regulation lineage

    ›Which regulations does the National Flood Insurance Program point back to in U.S. Code, and which agencies issued them?

    Traces a program from authorizing statute through codified law to issued regulation.

  • Crime

    ›How have FBI Crime Data Explorer county-level violent crime rates changed in major metros since 2018?

    Surfaces multi-year crime trends with source-native dimensions intact.

  • Population

    ›Which U.S. counties have the largest age-65-and-over share, and how has that changed since the 2020 Census?

    Combines Census decennial baselines with Population Estimates Program updates.

  • Health

    ›What does Medicare enrollment look like county-by-county, and how concentrated is it among the largest CMS regions?

    Joins CMS enrollment data to geographic concentration analysis.

How questions become golden

A candidate question is paired with an explicit agent plan, expected context types, and required tool groups. We run the plan, audit every claim against the structured facts it cites, and only promote the question to public after reviewer approval. The same contract that publishes the page also serves as a regression test for the agent.

Where the answer comes from

Question pages cite primary government records — never restated summaries. The agent retrieves from the same data sources exposed on the public catalog, and every citable claim points back to its source-release lineage.