How our newsroom works

Who this reporting is for

Our nonpartisan reporting puts working people's pocketbook issues first: pay, housing, groceries, healthcare, taxes, childcare and public services. We also explain which institution has authority and how a public decision connects to representation.

We aim to serve an urban, multiracial and multicultural audience without treating any community as politically uniform. Coverage addresses U.S. federal, state and local government. A ZIP code or household example is not evidence of someone's identity, beliefs, eligibility or expenses.

Sources before conclusions

We prioritize original bills, agency rules, budgets, government records and court orders, alongside independent reporting and community perspectives. A proposal is not an enacted law; an announced date does not prove implementation; a court ruling may be limited or stayed.

Linked evidence supports reported facts. Estimates, hypothetical household examples and editorial analysis are not measured outcomes or promises. Relevant examples may involve renters, hourly workers, commuters, caregivers, small businesses, families or retirees, rather than one assumed household.

What AI does, and what it cannot prove

The newly generated news workflow has three distinct stages:

1. Research current evidence. AI-assisted research gathers public records and reporting, explains practical effects and links its sources. Popularity signals are leads, not proof or public opinion.
2. Apply a separate automated copy desk. A second model-based research pass reviews claims, sources, policy status, civic authority and household effects. Daily national news and legislative briefings publish automatically after this review without a minimum confidence score. Scores and evidence gaps remain recorded and may be unresolved in published stories. Local paid news scans retain a 75% evidence-support threshold. Incomplete or malformed reviews and other processing failures preserve the previous edition.
3. Follow ongoing issues. A private pocketbook watchlist retains sourced milestones and next actions. OpenAI Decisions classifies research urgency from supplied metadata; it is not a government-decision feed or a fact-check. Every selected due issue remains eligible for fresh research, including routine, refused or unavailable classifications.

Automated review can still miss errors, misunderstand a source or produce unsupported conclusions. It is not a guarantee of accuracy, independent human fact-checking, legal advice or a voting recommendation. We do not claim that every briefing is reviewed by a person.

The internal confidence score is AI-assisted evidence coverage, not a calibrated probability that a story is accurate. Automatic publication can retain evidence gaps, including unsupported or uncertain claims. Publication does not change the original review findings or turn forecasts into facts.

Older editions retain their original publication dates, wording and media; this workflow does not mean they were retrospectively reviewed. Check the linked primary record for later amendments, deadlines, stays or other developments.