Editorial policy

How content on this site is produced, what software does and what humans decide, how everything is sourced, and how to get an error fixed.

Our standards

• Every number traces to a public source. Outage counts are the figures reported by the utilities and by ODIN, the U.S. Department of Energy's county outage data, and we never add outages a source didn't report; where a utility's area-by-area counts and its own total disagree, we reconcile the two before placing its outages in counties, by rules the methodology sets out. Where a displayed number that combines the utilities' own numbers with the federal county count differs from the county count alone, the page says so. Our sources page documents where our outage counts, weather alerts, utility territories and location lookups come from and how often the live ones update; the federal historical outage records some pages cite (EAGLE-I) are described on the methodology page.
• Safety and preparedness guides cite their sources. Each links the public sources its advice draws on (among them Ready.gov, the CDC, FoodSafety.gov, the FDA, the CPSC, the American Red Cross and HHS) and is general safety information, not medical or legal advice; our guide to how the forecast works cites our own accuracy, research and data sources pages.
• Predictions are graded in public. When a county outage that reached 200 or more customers ends, our county-wide estimate is re-run at each of its readings after its peak where the time was not withheld, from the outage's own recovery trend, and scored against when power actually came back, and the accuracy page publishes the results, misses included; the methodology sets out which outages are graded and how. The restoration times utilities post through the federal county outage data are graded by the same yardstick.
• No fabricated anything. No invented quotes, no fake author personas, no speculation presented as fact, and no content written to fill space. If we don't have the data, the page says so.

How content is produced — what software does, what humans decide

This site is data-driven and largely automated, and we think you should know exactly where the automation stops:

• Live pages are automated. The map, county/city/state pages, and utility pages render live utility and federal data on a ~15-minute cycle. No human writes "12,304 customers out" — the sources do.
• News articles are drafted by software, and publishing them is closed. Major outages auto-draft into structured articles from the live data, but none can be published — by a person or automatically — until our publishing tool shows the reviewer the full article as readers would see it, its sources and how it interprets them. Until October 6, 2026, a person could publish a draft after reading only its headline, summary and key figures; until October 1, 2026, drafts of outages above a size we set in advance published automatically, without a person reading them first, and those articles have been taken down. The publishing software also refuses outright a draft whose timeline is out of order, whose outage never reached 2,500 customers out, whose weather line is not plain text of 80 characters or fewer, or which has no county page to link to. The weather line is only ever taken from a National Weather Service alert for the county, and a posted restoration time that has already passed is shown as passed, never as a prediction.
• Restoration estimates are algorithmic — computed from each outage's own recovery trend, leaning on what past outages in the same place did while that trend is thin (method on the methodology page) — and clearly labeled as ours, separate from the utility's official time.
• Guides and pages about the site are written with AI software. The software that maintains this site drafts and corrects them and checks their claims against our code, our data and the sources they cite, under the direction of the site's operator. Each guide shows the date of the last such check of the whole page and, if its wording has changed since, the date of that change.

Authorship

Content is published under organizational authorship — "When's the Power Back? editorial" — with a visible reviewed or published date. We don't attach invented human bylines. Structured data identifies the organization as the author and publisher.

Corrections

If a number, claim, or citation is wrong, tell us via the contact form (choose "correction" or mention it in your message). What happens next:

• We check the claim against its source — the live data, the archived reading, or the cited authority.
• If it's wrong, we fix it. A corrected guide, or a page about the site like this one, shows the date of the correction as its updated date; its reviewed date moves only when the whole page is checked against its sources again. News publishing is closed, so no article is up to correct; when it reopens, a correction to a published article will be published in place with an updated timestamp distinct from the original publication time.
• If our data source itself was wrong (a utility's own outage map reporting an impossible number, for example), we correct or remove those readings in our records, so the affected pages and their outage history show the corrected figures.

Live outage numbers correct themselves on the next data refresh (~15 minutes); the corrections process above is for published articles, guides, and methodology claims.