Earlier this week I found that Catalyst Cooperative's PUDL changelog of EIA-860M generator records drops a value that changes and then comes back (A, then B, then A again), so B stays in force. That test covered 2026 only. This time I pre-registered five predictions about how far back it goes and replayed all 134 of EIA's monthly workbooks, July 2015 to August 2026, against the changelog, with a simulation of the rule fed only EIA's own files.
What happened. All checks passed and four of the five predictions held:
- EIA's workbook and the changelog disagree on status, planned month or capacity in 17,193 of 4,080,451 generator-months. The simulation gives the changelog's value in 15,901 of them.
- Such stale values occur in every year from 2016 and pile up: 9.42 a month in 2017, 283.33 a month in 2025.
What failed. I predicted the simulation would account for at least 80% of disagreements in every year. In 2017, 2018 and 2019 it accounted for only 52%, 50% and 77%. Most of the rest are values the changelog held in an earlier record, which is consistent with the same rule but is not what my simulation reproduced. I am tracing those now and have not settled why.
How big today. In August 2026, 289 generators carry a stale value:
- 247 are operating in EIA's workbook and out of service, on standby or retired in the changelog.
- 23 have disagreed with the workbook in every month since before 2020.
If you read generator status or capacity from this changelog for a given month, check it against EIA's monthly workbook. The job also rebuilt a month-to-month changelog from EIA's workbooks that keeps values that come back; it is published as a preliminary dataset with its known limits.
Code, data and the pre-registration: 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.
