One EIA Release Day Carried Two Reference Months: June Generation, July Capacity

On August 26, 2026, the U.S. Energy Information Administration posted two of its monthly electricity files. One of them described June. The other described July. Same agency, same release day, same broad heading of monthly electric power data.

Nothing about that is hidden, and it is not a mistake. Monthly electricity data is not one collection. It is several, each with its own frame of respondents, its own reporting frequency, and its own revision clock. A reader who treats the release date as the reference date imports that mismatch into everything built on top.

The useful question about an electricity statistic is rarely whether the number is large. It is what the number is a number of, and when it stops moving.

One Release Day, Two Reference Months

The EIA Electric Power Monthly landing page carries its own vintage in plain text: "Data for June 2026 Release Date: August 26, 2026 Next Release Date: September 24, 2026". The generation tables that went live in late August describe a month that had ended roughly eight weeks earlier.

The EIA Preliminary Monthly Electric Generator Inventory, built on Form EIA-860M, shows the same two dates: a release date of August 26, 2026 and a next release date of September 24, 2026. The most recent inventory file linked on that page is the July 2026 file, posted as july_generator2026.xlsx. The capacity inventory is one reference month ahead of the generation report that shipped beside it.

That gap is by design, because the two products answer different questions. EIA describes Form EIA-860M as a survey that "supplements the annual survey form EIA-860 data with monthly information that monitors the current status of existing and proposed generating units at electric power plants with 1 megawatt or greater of combined nameplate capacity". Monitoring current status is a near-real-time job. Measuring how many kilowatthours flowed in a given month waits on plant-level filings.

A sentence pairing the newest capacity figure with the newest generation figure is, without saying so, pairing July with June.

The One-Megawatt Line Is a Survey Boundary

Almost every trap in this dataset traces back to a single threshold that has nothing to do with physics. EIA's technical notes describe Form EIA-860 as "a mandatory annual census of all existing and planned electric generating facilities in the United States with a total generator nameplate capacity of 1 or more megawatts". A 2019 EIA explainer states the rule in the other direction: "EIA considers utility-scale generating facilities to be those where total generation capacity is one megawatt (MW) or greater."

Below that line, the machines still generate electricity. They simply stop being visible to the surveys that produce generation statistics. The same EIA Today in Energy explainer on utility-scale solar plant sizes says where they go instead: "Solar PV facilities with less than one megawatt in capacity are not included in EIA's surveys of electricity generators, but their aggregate capacities are included in the EIA's survey of electric power sales, revenue, and energy efficiency."

So a sub-megawatt rooftop array is not missing from the statistical system. It is routed into a different survey, which collects capacity rather than metered output. Generation for that tier has to be produced some other way, and EIA says plainly that it is: monthly generation from small-scale photovoltaic resources is "an estimation of the generation produced from PV solar resources and not the results of a data collection effort for generation directly, with the exception of 'Third Party Owned' or (TPO) solar installations which has direct data collection".

The reason is stated without euphemism in EIA's 2015 announcement of the series: "Because electric utilities do not necessarily know how much electricity is generated by rooftop PV on their distribution systems, generation from these systems must be estimated."

That is the definitional core. One published solar generation figure can be a metered sum, a modeled estimate, or a blend, and the difference is decided by a one-megawatt administrative line rather than by the technology.

The Monthly Total Rests on About 1,900 Filings

The second trap is quieter, because it sits inside the tier that looks solid. Plants above the one-megawatt line are not all reporting every month. The EIA Electric Power Monthly Technical Notes set out the split: "Data are collected from a monthly sample of approximately 1,900 plants, which includes a census of nuclear and pumped-storage hydroelectric plants. In addition approximately 4,050 plants, representing all other generators 1 MW or greater, are collected annually."

Roughly 5,950 plants sit in the frame. Fewer than a third of them file a monthly report. The rest are picked up once a year, which means the monthly figure attributed to them is an estimate rather than a sum of their own submissions. EIA treats it that way, and publishes the machinery for saying how uncertain it is: "there is approximately a 68 percent chance that the true total or mean is within one RSE of the estimated total or mean."

Appendix C of the Electric Power Monthly illustrates the scale of that error term with a worked example rather than a measurement, supposing "a net generation from coal value is estimated to be 1,507 million kilowatthours with an estimated RSE of 4.9 percent". That is a teaching example, not a reading for any month. What it establishes is that a relative standard error in the low single digits is the range the agency considers worth explaining.

Preliminary Does Not End Next Month

Revision lag is where the two traps compound. The intuitive model is that a preliminary monthly number gets corrected in the following month and settles. EIA's technical notes describe something slower: "All monthly survey data are first disseminated as preliminary. These data are revised after the prior year's data are finalized and are disseminated as revised preliminary."

The trigger is not the passage of a month but the finalization of the prior year. A monthly generation figure therefore holds preliminary status against an annual cycle, and the label it graduates into is revised preliminary rather than final.

The capacity side has a parallel clause, and a blunter one. EIA states that building the monthly inventory "sometimes requires the use of data submitted on the annual EIA-860 before it has been fully verified by EIA. For this reason, reported capacities are EIA's preliminary estimates of capacity for that month. Estimates will be corrected without specific acknowledgment or explanation in subsequent month's inventory."

Corrected without specific acknowledgment is the operative phrase for anyone maintaining a series. A value copied from one month's inventory and compared against the next can differ for two unlike reasons: capacity changed, or the earlier estimate was quietly replaced. The file does not distinguish them. The same structural problem appears in the piece on capacity utilization and its annually revised denominator and in the piece on benchmark revisions landing on a single month: a revision on an annual clock does not announce itself on a monthly chart.

The December 2015 Break in the Solar Series

Definitions also move, and the small-scale solar series carries a datable break. The EIA Form EIA-861M data page now publishes a small-scale estimate covering "capacity (in AC) and estimated generation from PV solar systems less than 1 megawatt in size". That did not always exist at monthly frequency.

EIA announced the addition on December 2, 2015: "Starting this month, the U.S. Energy Information Administration (EIA) is including monthly estimates of small-scale distributed solar PV capacity and generation by state and sector in EIA's Electric Power Monthly." The same notice describes what preceded it: "Although EIA has provided annual estimates of capacity and generation for small-scale PV systems at the national level in the Annual Energy Outlook for many years, it has not previously developed monthly state-level estimates for small-scale PV."

The base rate worth carrying from that episode is not about solar. It is that a series can gain a component without gaining a warning label. Any long comparison of monthly United States solar output spanning December 2015 sets a period with the monthly state-level small-scale estimate against one without it. Whether that changes a given conclusion depends on magnitudes, which a single stitched line will not show.

The sampling question compounds it once more. Respondents to Form EIA-861M are themselves selected from the larger group of respondents to the annual Form EIA-861, so the sub-megawatt tier is a sample-based capacity figure feeding a modeled generation figure.

Where This Reaches Capacity Readings

Capacity totals are the most commonly reused output of this system, and EIA's own Resource Advisory on understanding electricity generation capacity, dated July 24, 2025, sets two boundaries on how far they travel. On coverage: "EIA data on capacity are largely limited to utility-scale power plants." On the definition of the headline measure: "Net summer capacity is the maximum amount of power that generation equipment can supply to the grid at the time of summer peak demand."

Both matter for anything resembling a reserve margin calculation. A capacity total drawn from the generator inventory inherits the one-megawatt floor and the preliminary status at once. Net summer capacity is anchored to summer peak conditions by definition, so applying it unchanged to a winter question is a substitution the definition does not authorize. Earlier coverage of where the EIA electricity tables stop on prices and data center load hit the same wall from the demand side.

None of this makes the published numbers wrong. It makes them narrower than the sentences usually built from them.

The boundary rule in Figure 3 is worth stating in words, because a diagram that leaves an edge case unrouted is worse than no diagram. A unit at exactly 1.0 megawatt falls on the at-least side and is handled by Row 1 or Row 2. Every value below 1.0 megawatt, down to and including zero, routes to Row 3. There is no capacity value that matches no row, and none that matches two.

Three Ways the Next Release Can Read

Scenarios here are about interpretation risk rather than market direction, and they are not mutually exclusive.

Scenario A, the reference-month slip. A capacity figure from the newest generator inventory is set against a generation figure from the newest Electric Power Monthly, and the two describe different months. On August 26, 2026 that pairing was July against June. It is the most likely of the three, because both files are genuinely the newest of their kind and neither is labeled as mismatched with the other. The tell is a sentence that uses the word latest twice.

Scenario B, the tier blend. A solar total is read as measured output when part of it is a modeled estimate for systems below one megawatt. The size of the distortion depends on how large the small-scale tier is in the period discussed, a magnitude question this analysis does not resolve. The tell is a solar figure quoted without stating whether small-scale estimates are included.

Scenario C, the silent correction. A value stored from one monthly inventory is compared against the next month's file and the difference is attributed to real capacity change, when EIA has said estimates are corrected without specific acknowledgment. The counter-case deserves equal weight: most month-to-month differences presumably do reflect real additions and retirements, and treating every change as a revision artifact would be its own error. The file does not let an outside reader separate the two, so a difference of a few hundred megawatts should not be narrated as either without a second source.

What Would Invalidate This

The framing above rests on published documents rather than on inference, but several things would undercut it.

If EIA redesigned Form EIA-923 so that the monthly collection became a census of plants at or above one megawatt, the sampling argument behind Figure 2 would lapse and the monthly total would be a sum rather than an estimate. The plant counts used here are EIA's own approximations rather than fixed constants, and a later edition of the technical notes could state a different split.

If the one-megawatt threshold itself moved, Figure 3 would need rewriting rather than adjusting, because every row is defined against that line.

Most importantly, the reference-month offset is an observation from one release day, not a standing rule. A single release in which both products carry the same reference month would show the offset to be a scheduling artifact rather than a structural feature, and the August 26, 2026 pairing should not be generalized past what it is. The revision clauses are the durable part of this piece; the offset is the part most likely to change.

What to Watch Next Week

Two items matter in the days ahead: one is a date, the other is a status label. Both come from the pages themselves rather than from inference.

The Electric Power Monthly page lists its next release date as September 24, 2026. The Preliminary Monthly Electric Generator Inventory page lists the same next release date, September 24, 2026. The Form EIA-861M page states its own schedule differently, as "End of September 2026 for July 2026 data".

The check to run on September 24 is narrow: read the reference month printed on each product rather than assuming it advanced by one. The August 26 release showed a generation report for June alongside an inventory file for July, and whether that offset persists is an observation to make, not a forecast. The EIA-861M page is the one product that has already named its next reference month in writing, and it names July.

The second item is the revision clause rather than a date. Any figure pulled from the September release is preliminary under EIA's stated policy, and becomes revised preliminary only after the prior year is finalized. A number taken on September 24 should be recorded with that status attached.

Concrete Framework

A monitoring checklist for anyone reusing United States electricity statistics.

• Record the reference month, not the release date. Store both, and never let the release date stand in for the period described. The August 26, 2026 pairing of a June generation report with a July inventory file is the failure mode in its plainest form.
• Ask which side of one megawatt the figure sits on. At or above, it is inside the generator surveys. Below, including zero, it is outside them, and its generation is estimated rather than metered.
• Ask whether the plant filed monthly. About 1,900 plants in a frame of roughly 5,950 do. For the rest, the monthly value is an estimate carrying sampling error that EIA documents but does not print on every table.
• Attach a status label to every stored value. Preliminary and revised preliminary are different states, and the transition is triggered by the finalization of the prior year rather than by the next month's release.
• Do not diff two monthly inventory files and call the result capacity change. EIA states that estimates are corrected without specific acknowledgment, so the difference mixes real change with silent correction.
• Check whether a long series crosses December 2015. Monthly state-level small-scale solar estimates entered the Electric Power Monthly that month, and a comparison spanning it is not a comparison of like with like.
• Keep net summer capacity in its season. It is defined at the time of summer peak demand, so carrying it into a winter adequacy question changes the meaning of the number without changing its label.

None of these steps require anything proprietary. Each is answered by a sentence EIA has already published, usually in a technical note that ships with the data and is read far less often than the table beside it.