The projects only EIA's annual survey planned for 2017 were mostly ones the monthly survey first listed that year

In my last post, EIA's annual survey and its monthly survey (EIA-860M) listed different projects as planned for the same year: for 2017, 5,627 MW of the annual survey's 26,809 MW plan was not in January's EIA-860M plan, and 85% of it came online that year. A pre-registered test, run in a sealed job on PUDL's copies of EIA's data, now shows where those projects stood in EIA-860M. All four checks passed; 3 of 4 predictions held.

• Most of them were not in the monthly survey in January. Of the 5,627 MW, 3,530 MW (63%) was first listed in EIA-860M after January 2017, 617 MW of it as late as July-September. 1,436 MW was in January's report with a status that was not planned, 652 MW was planned there for a month outside 2017, and 9 MW never appeared under the same id. The prediction was at least 60%.
• They did not mostly arrive already being built. Of the 3,530 MW first listed after January, 61% was operating or under construction in its first EIA-860M record (336.6 MW operating, 1,812.5 MW building), and 1,200 MW was first listed at the earliest stages. The prediction was at least 80%, so it failed. Still, 81% of those 3,530 MW came online in 2017.
• It holds across years. The after-January share was at least 50% in all 8 years 2017-2024 (prediction: at least 6), and ranges from 58% (2025) to 80% (2023) over 2017-2025.
• From the other side, 62% of the 4,627 MW that EIA-860M added during 2017 without having listed it in January is in the annual survey's 2017 plan (prediction: at least 50%).
• Onshore wind: four of the five largest wind projects only the annual survey planned (Red Dirt, Thunder Ranch, Red Pine, Beaver Creek) were first listed in EIA-860M between February and July 2017. Beaver Creek (170 MW) first appeared at an early stage and still came online in 2017. None of the five is missing from EIA-860M.

So the annual survey's planned list for a year is partly a record of projects that became known during that year, not a forecast made in January, and a planned-against-built comparison that starts from it measures something different. Why the list includes them is not tested. Tables, the named projects, raw output and a test that checks every number: Where the annual survey's plan-only projects stood in EIA-860M

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