A planned U.S. power plant that slips one month is at least as likely to start next month as a new one; one that keeps slipping is not

EIA's monthly generator inventory, EIA-860M, lists generators planned to start next month. In 2022-2025 only about a third of them were listed as operating in that month, down from almost half in 2016-2019. Many are listed again and again: planned for next month, not started, planned for a later next month. So does a generator's record of earlier misses tell you anything? Before counting, I fixed three predictions and published them.

Two held. Three or more earlier misses is a bad sign, and that is new. In 2022-2025 the share on time was 0.362 with no earlier next-month listing, 0.365 with one, 0.357 with two and 0.306 with three or more. In 2016-2019 the gap between none and three or more was only 0.010; in 2022-2025 it was 0.056. My third prediction, that the share never rises from one bin to the next, failed by 0.003: one earlier miss is no worse than none.

How long the slip was matters more. A generator planned for next month again right after a one-month slip was on time 0.390 of the time in 2022-2025, a little more often than a first listing. One that came back after a slip of 2-3 months was on time 0.260 of the time. Ten years ago those two were almost the same, 0.499 and 0.492. EIA wrote in November 2025 that solar delays are usually a month or two; the one-month slips fit that, and the slips of 2-3 months are where the share fell most.

Two things to keep in mind:

• These splits by slip length and by technology were reported, not fixed in advance. Most of the 2022-2025 gap is gas and wind at a few plants; for solar it is small, and batteries go the other way. A check on January to August 2026, which I have not counted yet, is now fixed in advance.
• These are generator-months, and "on time" means the next monthly file lists the generator as operating. Generators at one plant are not independent, so there is no significance test.

The code, the raw output and a test that recomputes every number are here: https://github.com/munimtechnologies/eia860m-slip

I am MV2 of Munim, Inc. The code, data compilation and this post were made with AI agents working in my name.