ElasticTruth

Update, 30 October: the duplicates were deleted, not corrected

(Yesterday's post set out the timeline of the Sydney corrections. This is an update on what the files now show. link)

On 22 October 2026, the Bureau of Meteorology issued a statement on its temperature record. It said:

"We have conducted analysis of all underlying data examined in the report, which confirmed a small number of duplicated historical daily temperatures. These were 4 post-1910 months relevant to the Australian Climate Observations Reference Network – Surface Air Temperature (ACORN-SAT)."

"These account for fewer than 200 daily values from a dataset containing more than 7 million values."

"We have corrected these values in this year's update, completed on 19 October 2026."

The Bureau's letter to Andrew Wilkie MP, signed by Dr Stuart Minchin, says the same thing: four post-1910 months, fewer than 200 daily values.

What the public file now shows is not a correction of four months. It is a deletion of duplicate blocks at Sydney, and it is larger than the statement described.

What the file shows

The blocks identified in the audit at Sydney (station 066062), minimum temperature, were:

| Block | Dates | Length |

| 1 | July 1864 / July 1865 | 31 days |

| 2 | August 1867 / August 1907 | 31 days |

| 3 | May 1883 / May 1884 | 31 days |

| 4 | July 1914 / July 1915 | 31 days |

| 5 | November 1903 / November 1905 | 30 days |

| 6 | June 1943 / June 1944 | 30 days |

| 7 | December 1935 / December 1936 | 29 days in two pieces (14 + 15) |

| 8 | 1868 / 1865 | 29 days and 24 days |

That is eight blocks at a single station, totalling well over 200 days. The statement named four months and "fewer than 200 daily values" across the entire network. The Charleville maximum and minimum records add a further month.

The blocks are now gone from the file. They have not been replaced. There is no indication of which of the two copies was retained, or on what basis that was determined.

I cannot say the records have been "fixed".

I have not seen a replacement value. What I can say is that the duplicate blocks have been de-duplicated — one of each pair has been removed from the public file.

Whether the remaining values are correct, and whether they are the originals, is a separate question that the file does not answer. A correction replaces one value with another and leaves both on the record. A deletion removes one of the two and leaves nothing in its place. Those are different acts, and the statement described the second as the first.

The threshold: seven days

The deletion appears to have been applied down to a threshold of seven days. Blocks of seven days or longer have been removed. The seven-day block copied from January 1910 into February 1910 is still present.

That is worth stating plainly: if there is a threshold for de-duplication, it does not cover every block at that size. The January/February 1910 block is a seven-day copy across a month boundary, in the first year of the Bureau's standardised record, and it remains in the file.

The dates

- 19 October — the date the Bureau says the correction was completed.

- 22 October — the statement issued, naming four months and saying the correction was completed on 19 October.

- 26 October — the duplicates still present in the public file. Published on this blog. Dr Jennifer Marohasy writes to Media Watch identifying the August 1867 / August 1907 block as still present.

- 28 October — the duplicates still present. Raised in writing with three parliamentarians.

- 29 October — the duplicates removed from the public file. Media Watch writes to Dr Marohasy stating that the program "was correct when we went to air" and, in the same letter, that the raw data "was also in the process of being fixed, and I believe, should now be done."

The statement said the correction was completed on 19 October. The file on 28 October still contained the duplicates. The duplicates were removed on 29 October.

The questions

I am setting out what the files show, and asking the Bureau four questions:

1. On what date was each block removed from the public file?

2. For each pair, which copy was retained, and on what basis was that determined?

3. Why did the 22 October statement name four months, when the deletion covers at least eight blocks at a single station and more than 200 days?

4. What is the threshold for de-duplication, and why is the January/February 1910 block still present?

The dates are checkable. The files are public. Anyone can verify the sequence.

What this changes

The dispute is no longer about whether the duplicates existed. They existed, and they are now gone.

The dispute is about three things:

- Scale. The statement described four months. The deletion covers at least eight blocks at one station.

- Method. The statement said the values were "corrected." The file shows duplicate blocks removed, with no replacement value and no indication of which copy was retained.

- Timeline. The statement said the work was completed on 19 October. The file shows it was still to be done on 28 October and was done on 29 October.

Those three points require the Bureau to show its work. That is what the statement has not done, and what the file does not show.