75% of EIA-923 Page-4 'Zero Output 2025' Generators Are Reporting Artifacts, Not Stranded Assets

USE CASE. Grid planners (RTO/ISO long-term planning teams at PJM, MISO, ERCOT, CAISO, ISO-NE), state public utility commissions running integrated resource plans, decarbonization modelers (NREL, Princeton ZERO Lab, Resources for the Future), and reliability analysts at NERC need to identify generators that are de facto retired before they appear in the official EIA-860M Retired sheet. The natural query against EIA-923 + EIA-860M is 'Operating in 860M, zero output for 2025 in 923'. RESULT 1 (the apparent Page-4 finding). Filtering EIA-923 'Page 4 Generator Data' (the per-generator monthly time series) for generators with all reported 2025 months below 50 MWh and at least 6 months reported, then joining to EIA-860M Operating, yields 122 generators totaling 10,180 MW of nameplate capacity. The top 5 by capacity: Manatee Units 1 and 2 (FL, 809 MW each, 1976-77 vintage natural gas steam turbines, FPL-owned), Martin Lake Unit 1 (TX, 800 MW, sub-bituminous coal steam, Vistra-owned, 1977 vintage), and Greensville County Power Station ST01 (VA, 609.9 MW, natural gas combined cycle, 2018 vintage, Dominion Energy-owned). The Manatee and Martin Lake entries are plausible (older steam units in deep mothball or unplanned extended outage), but the Greensville County entry is suspicious — combined-cycle plants from 2018 do not typically sit idle for an entire year, and Greensville is one of Dominion's largest CCGT facilities. RESULT 2 (the structural artifact). Cross-checking each Page 4 'zero output' generator against the same plant's Page 1 (Generation and Fuel Data) totals reveals that 92 of 122 (75%) are reporting-level artifacts: the operator reports at the PLANT level on Page 1 (with non-zero monthly netgen) but does not file individual GENERATOR level rows on Page 4, leaving the Page 4 cells empty. Greensville County is the canonical example: Page 4 shows three CT/ST units at zero, but Page 1 shows the plant total at 22-63 GWh per month — completely normal CCGT operation. The same applies to all six Virginia Dominion plants (Greensville, Brunswick County, Warren County, Bear Garden, Possum Point ST6) and to Manatee, Martin Lake, and the Bethlehem Energy Center NY. Once the Page 1 cross-check filter is applied, the truly stranded set drops from 122 generators / 10,180 MW to 30 generators / 357 MW. RESULT 3 (the truly stranded list). The 30 verified stranded generators are dominated by small industrial cogeneration and institutional plants. Top 10 by capacity: International Paper Georgetown Mill SC GEN3 (37.7 MW biomass, 1984), Wyandotte MI Unit 7 (32.0 MW NG steam, 1986), Georgia-Pacific Crossett AR GEN4 (28.0 MW NG steam, 1962), International Paper Georgetown Mill GEN2 (27.4 MW biomass), International Paper Georgetown Mill GEN1 (25.3 MW black liquor biomass, 1966), CSUCI Site Authority CA GEN1 (21.5 MW NG combined cycle, 1988 — Cal State Channel Islands campus plant), Eagle Point Energy Center ME GEN1 (21.2 MW MSW, 1987), Indeck Jonesboro Energy Center ME GEN1 (20.2 MW biomass, 1987), Camden SC GEN3 (17.5 MW NG steam, 1993), University of Colorado GT2 (15.0 MW NG combined cycle, 1992). The structural pattern: most truly-stranded generators are small (< 40 MW), built between 1948-2008, and at industrial (paper mill cogeneration) or institutional (university campus power plant) sites where the host has likely transitioned to grid purchases. The largest cluster is International Paper's Georgetown Mill in South Carolina with three generators totaling 90 MW all showing zero output for 2025 — consistent with the mill having discontinued self-generation entirely in 2025. The University of Iowa Main Power Plant has three generators totaling 30 MW showing zero output, two of which are vintage 2022 (recent installations now sitting idle, unusual). STRUCTURAL READING. EIA-923 has TWO distinct net-generation reporting paths: Page 1 (plant-prime-mover-fuel level) is mandatory and has high coverage; Page 4 (generator level) is filed by a subset of operators and is structurally incomplete. A grid analyst running a stranded-asset query against Page 4 alone gets a 4x false-positive rate dominated by major-utility CCGT plants that are operating normally. The right interpretive practice is to use Page 1 plant totals as the gating filter, then drop into Page 4 only for plants whose Page 1 total is also near zero. CAVEATS. (1) The 50 MWh per-month tolerance is conservative — some truly-zero generators may have parasitic loads slightly above the threshold and be missed. (2) The Page 1 join is at the PLANT level; plants with multiple generators where some are stranded and others operating won't have their stranded subset isolated by this filter (the filter is conservative — it returns true positives but may miss partial-stranded plants). (3) The 30-generator verified list does not include plants that file ONLY on Page 1 with no Page 4 rows at all; those plants would appear as 'no Page 4 entry' rather than 'Page 4 entry with zero output' and are excluded by construction. The true stranded universe is therefore at least 30 generators / 357 MW and likely larger.

EIA, the federal energy statistics agency, publishes monthly reports from every US power plant. Two of these reports give complementary information: one (EIA-860M) lists every generator and whether it's currently 'Operating' or 'Retired'; the other (EIA-923) lists how much electricity each generator actually produced each month. If you cross-reference them, you can ask a useful question: which power generators are listed as Operating but actually produced zero electricity all year? Those are de facto stranded assets — power plants that haven't been formally retired but aren't doing anything anymore. I downloaded both files this morning and ran the join. The first answer looked striking: 122 generators totaling 10,180 megawatts (about 10 nuclear reactors' worth of nameplate) appeared to be 'Operating but zero output' for all of 2025. The list included things like Florida's Manatee plant (1,600 MW), Texas's Martin Lake coal plant (800 MW), and several large Virginia combined-cycle gas plants. But that didn't pass a sanity check — combined-cycle gas plants from 2018 don't typically sit idle for a year. So I checked: EIA-923 has TWO different sheets reporting net generation. Sheet 1 ('Page 1') reports at the plant level, and Sheet 4 ('Page 4') reports at the individual generator level. Sheet 4 is filed by a smaller number of operators. When I cross-checked the supposedly-stranded generators against Page 1 plant totals, 92 of 122 (75%) of them were operating normally on Page 1 — the operator just doesn't bother filing the generator-level breakdown. Greensville County in Virginia, for example, shows 22-63 gigawatt-hours of monthly output on Page 1 but completely blank Page 4 rows. So the original list was a 4x false positive: anyone using EIA-923 Page 4 alone to find stranded assets would conclude there's 10 GW of phantom retirements, when 75% of those are just operators choosing not to file the generator-level detail. After the Page 1 cross-check, the actual truly stranded list is 30 generators totaling 357 megawatts. They're mostly small industrial cogeneration plants (paper mills, biomass) and university campus power plants — places where the host facility has apparently switched to grid purchases. The biggest single cluster is International Paper's Georgetown Mill in South Carolina with three biomass generators totaling 90 MW, all showing zero 2025 output. Why this matters: grid planners, state utility commissions, and decarbonization analysts use EIA-923 + EIA-860M to identify capacity that's de facto off the system. Using only Page 4 gives them a 4x false positive rate dominated by big-utility CCGT plants. The fix is simple: use Page 1 as the gating filter. The fact that EIA itself doesn't publish a stranded-asset list with this cross-check is a real gap.

International Paper Georgetown Mill SC GEN3 37.7 MW biomass 1984 · Wyandotte MI Unit 7 32.0 MW NG steam 1986 · Georgia-Pacific Crossett AR GEN4 28.0 MW NG steam 1962 · International Paper Georgetown Mill GEN2 27.4 MW biomass 1984 · International Paper Georgetown Mill GEN1 25.3 MW black liquor biomass 1966 · CSUCI Site Authority CA GEN1 21.5 MW NG combined cycle 1988 · Eagle Point Energy Center ME GEN1 21.2 MW MSW 1987 · Indeck Jonesboro Energy Center ME GEN1 20.2 MW biomass 1987 · Camden SC GEN3 17.5 MW NG steam 1993 · University of Colorado GT2 15.0 MW NG combined cycle 1992

EIA-923 Page 4 zero-output 2025 generators (Operating in EIA-860M): 122 generators / 10,180 MW · After Page 1 plant-total cross-check: 30 generators / 357 MW are truly stranded · 92 generators / 9,823 MW are reporting artifacts (operator files at plant level on Page 1, leaves generator-level Page 4 rows empty) · false positive rate: 75% by count, 96.5% by capacity · canonical artifact example: Greensville County Power Station VA (3 CCGT units showing zero on Page 4, plant total 473,877 MWh in 2025 on Page 1)

2,628 generator rows on EIA-923 Page 4 · 27,718 Operating generators in EIA-860M · 140 Page 4 generators with all reported 2025 months below 50 MWh · 122 of those confirmed Operating in 860M · 30 of those have plant-level Page 1 also at or below tolerance · stranded by technology: NG Steam 17 units 153 MW, Biomass 5 units 121 MW, NG CCGT 4 units 44 MW, MSW 2 units 26 MW