When's the Power Back?

How accurate are these estimates?

Our independent, county-wide restoration estimates, graded against what actually happened. Unofficial — always follow your utility's guidance.

At a glance

Across 600 scored outages since Aug 17, 2026, median absolute error 1.9 h (54% within 2 h). For 52 time-matched cases, our median error was 1.4 h versus the utility's 1.5 h; ours was lower in 50% of them.

7,956
outages scored all-time
of 15,299 tracked (52%)
1.9 h
median absolute error
across the 600 most recent scored
2.9 h
first forecast error
made early in an outage (n=600)
0.9 h
final forecast error
made just before power returned
1.5 h
utility's own estimate
median error at the same readings (n=52); ours: 1.4 h

Every error figure is a median absolute error — half the estimates landed within that many hours of the actual restoration. The utility comparison is time-matched — both graded on the same readings against the same restoration.

How close, across the 600 most recent scored outages: 36% of estimates landed within 1 hour of the actual restoration, 54% within 2 hours, 69% within 4. Typical direction of the miss: 0.1 h early — we tend to call restoration sooner than it happens.

Utilities pushed their posted time back at least once in 81% of 59 outages with a posted time (median 4× when they did).

Median error by time into the outage: 2.8 h at 1 h in (n=567) · 2.9 h at 3 h in (n=429) · 2.5 h at 6 h in (n=293). Each figure is a different set of outages — only those that lasted at least that long — so read them as separate cohorts, not one outage sharpening over time (a survivor effect: longer outages dominate the later horizons).

Median error by outage size: 1.9 h for 500-2k customers out (n=244) · 1.7 h for <500 customers out (n=239) · 1.9 h for 2k-10k customers out (n=105) · 6.9 h for 10k-50k customers out (n=10).

Data updated Aug 19, 8:00 PM EDT · methodology 2026-07-29.v2 · new events land ~3 h after restoration holds. Preliminary until the sample grows. Every statistic above except the all-time count is computed over the 600 most recently scored outages — the window the published ledger retains — not all 7,956. The gap between tracked and scored is outages where unscored events made no forecast — too small to publish an estimate for, or restored before one became dependable — so there is nothing to grade.

Related: how often utilities push back their posted restoration times — same ledger, utilities graded.

Method

While an outage is active, the model watches how many customers are out and how fast the number is falling, projects the rest of the recovery forward with the pace easing off as it drags on, since crews clear the easy faults first — and leans on weather and past outages in that area. Recalculated ~every 15 min; independent of the utility. When a utility also posts a time, both are shown, labeled separately.

How we grade

Only outages that have fully ended are graded, and we never score an estimate against another estimate. When an outage's recovery never settled, the site showed "no clear recovery yet" — a deliberate non-answer, so nothing to grade (that's why "scored" is smaller than "tracked").

Is accuracy improving as data accumulates?

Error here is normalized by each outage's length — a 5-hour miss on a two-day outage isn't the same as on a five-hour one. Raw hours-of-error mostly tracks how severe that week's storms were, so this is the cleaner way to watch the estimator itself improve; the typical outage column shows whether a cohort was simply harder.

Prediction error by cohort of close date, normalized by outage length.
Cohort (by close date)OutagesError ÷ lengthMedian errorTypical outage
Earliest (8/17–8/18)19924%±1.9 h~8 h
Middle (8/18–8/19)19923%±2.3 h~10 h
Latest (8/19–8/19)19921%±1.6 h~8 h

Roughly flat so far. Preliminary — a real trend needs weeks of varied storms.

What each number measures

What each number measures — the published definitions for this methodology version (2026-07-29.v2), quoted from the metric contract itself.
our median absolute error (whole event)
Median absolute error, in hours, of every estimate we published across a completed outage, against its observed effective restoration (count first at/below 10% of peak and holding). Median across events.
posted-time median absolute error (as recorded)
Median absolute error, in hours, of the posted restoration times we recorded — a restoration time carried by the national county feed (the only posted-time source the event series records) — over the readings where one was present while meaningfully out, against the same observed restoration. Median across those events. Events where the utility posted a time that never reached that feed are NOT counted.
our median absolute error on the same posted-time events
independentWholeEventMae restricted to the events counted by utilityPostedMae — same event population, whole-event alignment (not the same readings). It inherits that metric's posted-time source limitation exactly.
time-matched head-to-head (same readings)
Both graded at the identical readings where a recorded posted time — a restoration time carried by the national county feed (the only posted-time source the event series records) — AND our estimator both produced a value: the strictest like-for-like. Reported as a pair (posted vs ours) with its own n.
posted time moved later at least once
Share of events carrying a recorded posted time — a restoration time carried by the national county feed (the only posted-time source the event series records) — where that time moved later by >30 minutes at least once before restoration. Earlier moves are not counted, and events whose posted time never reached that feed are not in the denominator.
average later revisions per posted-time event
Mean count of >30-minute later moves of the recorded posted time per event carrying one — a restoration time carried by the national county feed (the only posted-time source the event series records) — counting 0 when it held.

Recent examples from the scored sample

The newest 10 scored events (most recent first — not a curated or representative sample); the aggregate statistics above cover the full sample. Records of individual outages and downloadable datasets aren't part of the public site; contact us about future API access.

The newest scored outage events, with our median prediction error and the utility's where one was posted.
CountyClosedPeak outRestore tookOur errorUtility's error
Washoe, NV2026-08-192913.8 h±0.3 h
Susquehanna, PA2026-08-194827.5 h±4.1 h
Jefferson, LA2026-08-191,4176.5 h±1.6 h
870232026-08-191,1108.2 h±0.7 h
Sumter, AL2026-08-1980520.3 h±4.1 h
870602026-08-192118 h±27.4 h
870572026-08-1932215.7 h±1.8 h
Whatcom, WA2026-08-1922919 h±1 h
utility ±3 h
±3 h
Douglas, OR2026-08-192196 h±0.5 h
utility ±0.3 h
±0.3 h
Muskegon, MI2026-08-1924418.5 h±0.7 h
utility ±3 h
±3 h

Newest 10 of 600 scored. Error = median difference between predicted and actual hours-to-restore across the estimates published during that outage. "—" in the utility column means no official restoration time was posted.