How our independent restoration forecast works
What it is
Live outage status tells you what is out. Our restoration forecast estimates when power may come back — including when your utility has not posted an official time. It is independent and unofficial: computed from public outage data, never the utility's commitment.
How it's computed
We analyze how quickly customers are being restored, whether the outage is still growing, outage size, and similar past outages in the same area, and a county-wide estimate also takes the weather into account. The estimate recalculates with each new reading: about every 15 minutes for a county, and usually more often where we read the utility's own outage map. Where enough graded history exists, a county-wide time also shows a range grounded in our own past accuracy (for example "likely 2–5 hr") instead of false precision.
When we don't show a time
Honesty beats filling the card. You'll see "Outage detected — gathering data to estimate restoration" when there aren't enough readings yet, a note that too few customers are out for a meaningful county-wide estimate when an outage is very small, and "No clear recovery trend yet" while outages aren't steadily declining. A falling count can reverse, so for a county-wide estimate, until there is evidence the peak has passed, the card says the outage "may still be growing" instead of quoting a time to restore. City and neighborhood estimates don't have that check: they can quote a time as soon as the count is clearly falling. We never invent a return time.
Official ETA vs our forecast
When the utility posts its own restoration time, we show both, side by side, labeled separately — the official ETA is never silently replaced. The two answer different questions: the utility knows its crews; our forecast watches what the outage is actually doing.
How it's graded
When a county outage that reached 200 or more customers ends, we grade our county-wide estimate against when most of its customers got power back: the "90% restored" mark, a tenth of the outage's peak. At each reading after the peak while the count is above that mark — skipping those where the peak check or a delayed federal count withheld the time, or the pace showed no clear recovery — we re-run the live projection, aimed at the mark and using the outage's own peak, and score it against when the county's outage count next fell to the mark. That grades the live pace alone: a time the card also weighed against past outages is graded without that weighting, and a time past 48 hours, which the card shows only as "restoring slowly", is graded too.
We publish the results, including misses: how many hours off we typically are (with sample sizes), accuracy by outage size, and a side-by-side comparison with the posted times that reach us through the federal county outage data (not the times on utilities' own outage maps). City and neighborhood estimates are not graded yet. The reliability research grades those posted times by the same ground truth.
What the forecast is not
It is not a commitment — no one outside your utility dispatches crews, and we never present our number as theirs. It is not street-by-street: a county estimate describes the county's recovery as a whole, and city or neighborhood estimates (where a utility publishes that detail) describe those areas. And it is never a guess dressed as data: every status the card can show — gathering, growing, holding, restoring, a range — corresponds to a specific measured condition, and the vocabulary is deliberately fixed so the same words always mean the same thing.
The statuses you'll see, explained
"Gathering data" — not enough readings yet (a fresh outage needs a second data point). "Still growing" — the affected count is rising; quoting a restoration time now would be premature, so we say so instead. "Holding steady" (county pages call this "No clear recovery trend yet") — numbers aren't moving enough to trend either way. "~N hr to restore" — a live rate exists (and, for a county-wide estimate, the peak looks behind us).
"Likely X–Y hr" — a range beside a county-wide time, taken from our own graded record. When a county outage ends, each of its graded projections of 9 minutes or more gives a ratio: the time the count actually took to reach the "90% restored" mark, divided by the time projected. The ratios are pooled by how many readings each projection was made from, each pool keeping its most recent 400, and the range is the live time multiplied by the middle half of its pool's ratios (the 25th to the 75th percentile). A pool shows no range until 50 ratios back it.
Why an estimate can disappear
Data can stop arriving. When the slower nationwide county tally we read stops updating, the card says the count is delayed and when it last updated, and shows no estimate until fresh numbers arrive: a count that has stopped updating is never read as restoration. The estimate returns when readings do. A county dropping off that tally altogether is different: the tally lists the counties that have outages, so that is how it reports one ending.
The count can also move between estimates: where we read the utilities' own outage maps, a county's count is refreshed more often than its estimate is recalculated. When the newer count no longer fits the estimate (a recovery that stalled, or an outage that grew), the card says "The count just changed", gives the count the last estimate was worked out for, and shows no time until the next update re-estimates from the new count, about every 15 minutes.
How to actually use it
Tie decisions to thresholds, not vibes: at what estimate do you move food to a cooler (the 4-hour window makes that one easy), at what point does the household relocate (heat, cold, or medical needs set it), when is the generator worth starting? Use the far edge of a range for anything with a hard deadline, and watch the trend, not just the number — an estimate that keeps moving later is itself information.
Limits to keep in mind
County-wide estimates cover a whole county — your street can come back earlier or later. Coverage depends on which utilities report into public sources, and missing data is not the same as zero outages. In any emergency, follow your utility and 911 — never this site.