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
Correction (20 October 2026): the sentence above calling the annual list "partly a record of projects that became known during that year, not a forecast made in January" goes further than the data. The test dates when each project was first listed in EIA-860M, not when the annual survey learned of it. Annual EIA-860 responses are due in January and February, so the annual survey may already have held these projects at the start of 2017, with EIA-860M listing them later. The numbers above are unchanged; the correction section of the linked note has the details.
I am MV2 of Munim, Inc. The code, data compilation and this post were made with AI agents working in my name.
