On the morning of Friday 7 August, between 08:34:03 and 08:35:37 Greenwich Mean Time, the Department for Energy Security and Net Zero republished seven pages of fuel poverty statistics. A collection page followed sixteen seconds later, making eight pages in 110 seconds. Each carries the same change note: “a calculation error on benefit receipt”. What was corrected was six years of tables, from 2020 data to 2025 data, and the current annual report.

Extracting the text of the 98-page report as it stands and diffing it against the version the Internet Archive captured in March returns exactly two substantive differences in the whole document. The first is a change in the wording of a new affordability indicator’s formula. The second is one number, on page 47: “The share of fuel poor households receiving benefits rises to 61.0% if disability benefits are included.” In March that sentence read 67.3%.

The spreadsheets behind it show where the error lived. Each detailed-tables workbook has 39 sheets, and comparing every cell of the pre-correction files against the live ones finds exactly 21 differing cells in each — all of them in the Table 34 sheet, plus a date and a note. Table 34a, which counts fuel-poor households on benefits excluding disability benefits, is unchanged in all six years. Table 34b, the version including them, moved every time: 72.968 to 68.121 per cent for 2022, 68.254 to 62.374 for 2023, 68.916 to 62.697 for 2024, and 67.323 to 60.951 for 2025.

Because 34a did not move and 34b did, the difference between them isolates the households that qualify only through a disability benefit — a figure the department does not itself publish. For 2025 data it falls from about 229,000 households to 78,268. For 2024, from 242,000 to 87,866; for 2023, from 258,000 to 93,728; for 2022, from 240,000 to 85,856. That is an overstatement of roughly 2.8 times in each of the four years whose original workbook can still be retrieved. The 2022 annual report, which was never reissued, prints enough to reconstruct 2020 as well: about 209,000 against a corrected 95,594, or 2.2 times.

The corrected cells carry their own signature. In every one of the six republished workbooks, the Table 34 sheet is the only one of 39 with values printed to more than three decimal places — sixteen such cells in the file for 2025 data, ten in each of the others. Neither workbook DESNZ left alone contains a single one. The recalculated figures were pasted in raw.

The error had been in the public record for a long time. The oldest affected publication went up on 24 February 2022; the correction landed 1,624 days later. The current report had carried it for 133 days.

The republished report does not say any of this. Its cover still reads “26th March 2026”, it still carries the Accredited Official Statistics badge, and a search of all 98 pages finds no correction notice, no erratum and no revisions statement. The only record is the change note on the GOV.UK page and a line inside each workbook. There is no methodological note explaining what the calculation error was.

It also did not finish the job. The detailed tables published in 2025 for 2024 data are still live, were not in the 7 August batch, and still show 68.72 per cent in the same cell — the uncorrected value, matching the 68.7 per cent the 2025 annual report still prints. Three earlier annual reports print the same figure for their own vintages: 75.4 per cent for 2020 data, 72.1 for 2022, 71.0 for 2023. None was reissued, and the change history on each of those pages contains nothing but “First published.” The department corrected the newest file for each data year and left the superseded ones standing.

Two things did not change. None of the headline fuel poverty numbers moved — the proportion of English households in fuel poverty, the fuel poverty gap and the new affordability indicators are identical in both versions. And the series is not continuous across the correction: the basket of benefits behind Tables 34a and 34b was itself redefined between the 2024 and 2025 publications, so the corrected 2020 to 2022 figures and the corrected 2023 to 2025 figures are not measuring quite the same thing.