Four outages in Ontario County, and how our restoration estimate did

What this page is

These are four real outages in Ontario County, New York, reconstructed from the readings we archived when each one ended: a storm, a quick outage, a miss, and one that ended too fast to grade. They were chosen to show the range of what happens to our estimate, not to flatter it.

The page was put together by the AI software that maintains this site, from the public outage archive behind the county's history panel; no person has reviewed it yet. All times are Eastern.

What is graded is not exactly what a page showed at the time. After an outage ends, we re-run the pace estimate at each reading after the peak and score it against when the county's count first fell to a tenth of the peak, the point where about 90% of the customers were back. The live card adds checks of its own, such as waiting for evidence that the peak has passed, so it can show less than the re-run does. The methodology has the full rule, and the accuracy page the totals.

A storm: the early pace said hours, it took most of a day

The outage began on July 18 at 4:00 PM ET, in weather the archive classes as wind. It peaked at 4,057 customers out at 8:16 PM, and first fell to a tenth of that (405 or fewer) on July 19 at 2:15 PM ET: about 18 hours after the peak.

The count fell fast at first, then slowly. Replaying the grader over the readings the archive keeps: at 10:45 PM, with 1,741 out, the pace projected about 2 hours to that mark; it took about 15 hours. The grade recorded when the outage ended, over all its readings, is an average miss of 8 hours. In this outage the estimate ran too short: the pace early in the night did not hold.

A quick outage: close, on little evidence

The outage began on September 25 at 2:45 PM ET. It peaked at 478 customers out at 4:30 PM and first fell to 47 or fewer at 5:15 PM, about 45 minutes later.

The recorded grade is an average miss of 12 minutes. That reads well, but it rests on very little: by the first reading after the peak only 51 customers were still out, and only 2 readings above the mark could be graded. A close score on a small, fast outage says less than one on a long outage.

A miss by 2.8 days: a slow trickle, then nearly everyone at once

On October 8 at 1:45 PM ET, the count stood at 799 customers out. 30 minutes later it was 791: 8 customers back. At that pace the re-run projected about 44 hours to reach 79 or fewer, and, as the count held at 791, about 3.7 days at 2:45 PM.

Then, at 3:00 PM, the count fell to 2: nearly everyone came back at once, about an hour after the peak. The recorded grade is an average miss of 2.8 days. A pace estimate projects the pace it sees. It cannot see a repair that brings back hundreds of customers in one step, and here the pace it saw was nearly zero. This page does not reconstruct what the county's card showed at those readings: the card holds back a time in some conditions the re-run does not.

Too fast to grade: 1,093 customers, most back within 15 minutes

On September 28 at 1:45 PM ET, the count stood at 1,093 customers out. At the next reading, 2:00 PM, it was 12: already below a tenth of the peak.

The grade starts at the readings after the peak while the count is still above that mark, and this outage left none, so it was not scored. The accuracy record counts an outage like this as not scored, and says how many were, rather than calling every outage graded.

What the four show

• A pace estimate projects the pace it sees. When restoration slows, it runs short, as in the storm; when it jumps, it runs long, as in the miss.
• A grade is an average over readings. A quick outage gives the grader few readings, so a close score there is weak evidence.
• Some outages cannot be graded, and the accuracy record counts them as not scored instead of leaving them out silently.
• When your utility posts a restoration time, read both. The utility knows where its crews are; our estimate only sees the count.