ElasticTruth

BoM claim of fixing problems from the Audit is falsified by it's own data. 

Who's Fooling Who? The BoM's Statement says it found no further duplicate months. The files say otherwise.

The Georgetown example: 34 days at exactly 37.8°C

At Georgetown (station 030018), between 1 January and 8 February 1962, the maximum temperature was recorded as exactly 37.8°C on 34 of 39 consecutive days, at 0.1°C precision. Every value is flagged Quality = Y by the Bureau's own quality control.

The five days that are not 37.8 are 38.9, 38.3, 38.9, 38.3 and 38.3 — every one of them a single-day increment above the base value, followed by a return to 37.8.

This is not a transcription error. A transcription error is one wrong number. This is thirty-four days, across a month boundary, all landing on the same value to one decimal place, interrupted only by single days. 

It is the signature of a value being carried forward with occasional edits — the same mechanism that produces the 3+1+3 and 5+1+5 patterns documented across the network in the audit, and the same mechanism that produces the decimal-loss windows between 1995 and 2011.

Two things follow. 

First, the BOM's explanation of "isolated transcription errors from pre-1946 handwritten records" does not reach this file: it is 1962, it is post-formation, and it is flagged Quality = Y. 

Second, a station whose maximum is carried forward for 34 of 39 days is a station whose variance, extremes and percentile distributions have collapsed. 

That collapse is not corrected by homogenisation; it is fed into it.

The file is public. Anyone can open it and count. That is the point.

Sydney: The 1935/1936 Seam

December 1935, copied into December 1936 — with a seam in the middle

At Sydney (station 066062), the December 1935 and December 1936 minimum temperature records are not two months of observations. They are one block of data, joined to another block, with a visible stitch.

Read the two months side by side and the structure is unmistakable:

- December 1–14, 1935 and 1936: fourteen days, identical to 0.1°C.

- December 15–16: present only in 1936. The 1935 column is empty for those two days.

- December 17: both columns show 20. In the 1935 record, that value carries no days-of-accumulation flag — the flag that tells the reader whether the observation is complete. It is the one malformed record in the month.

- December 18–31, 1935 and 1936: fourteen more days, identical to 0.1°C.

Every value in both months is flagged Quality = Y by the Bureau's own quality control.

Why this is not a transcription error

A transcription error is one wrong number. It is not fourteen identical days, a two-day gap, a malformed record at the join, and fourteen more identical days. That is the shape of a block splice — two blocks of data joined together, with the join point left visible in the file. The two-day insert and the missing accumulation flag are the stitch marks.

The distinction matters because the Bureau's statement explains the duplicates as "isolated transcription errors" from pre-1946 handwritten records. This is 1935 and 1936 — inside that window — and it is not an isolated error. It is a join.

What the Bureau's statement says

The Bureau's statement of 22 October says it has "detected no further duplicate months in the entire bureau network other than those quoted in the report and then confirmed by us."

This is a further duplicate month. It is inside the pre-1946 window the Bureau cites. It is flagged Quality = Y. And it has a seam in it, which means the duplication was not a single copy operation but a join of two blocks — a different mechanism from the one the Bureau describes.

Why it matters

A block splice is worse than a copy, because a copy can be explained as a data-entry slip and a splice cannot. A splice means that two blocks of data were combined, and the join was not checked. The two-day gap and the malformed record are evidence that the join was not checked. If the join was not checked in 1935/1936, the question is how many other joins were not checked.

The file is public. Anyone can open it and see the stitch.

June 1943. June 1944. Thirty days. Identical.

The Bureau's statement of 22 October says:

"Further investigation has detected no further duplicate months in the entire bureau network other than those quoted in the report and then confirmed by us."

Here is a further duplicate month.

At Sydney (station 066062), all thirty days of June 1943 are replicated exactly in June 1944, to 0.1°C. Every value in both months is flagged Quality = Y by the Bureau's own quality control.

Thirty days. Not one digit different. Two complete months of daily observations cannot both be original. One is a copy of the other. The quality flag cannot be correct for both.

This is not a transcription error. A transcription error is one wrong number, not thirty identical numbers in the same order a year apart.

This is not pre-1910, and it cannot be set aside as outside the ACORN-SAT window. It is 1943 and 1944 — inside the pre-1946 window the Bureau itself cites as the source of the duplicates. The "pre-1946" explanation does not dispose of this file. It includes it.

This is not among the four months the Bureau says it corrected. It is still in the public record, still flagged Quality = Y.

The Bureau's statement contains a sentence. This file contradicts it. Anyone can open both and see.