3 September 2026
An improvement that sounds self-evident, and a way to test it before it ships.
The obvious improvement
In the years with no documented price, the price skeleton is weighted with headline CPI. For a food item, the Food group looks like the obviously better choice.
Its rates are not in any database for that period. They came out of the monthly statistical bulletins of ELSTAT, issue by issue, scans with no text layer. 22 years of them.
The test
There are two documented souvlaki prices in the drachma period, from two independent price-control orders, 1977 and 1989. Take the first, carry it forward to the second with each index, and see which projection lands closer.
Headline CPI misses by 3.4%. Food misses by 5.9%. Nearly double the error, from the series we assumed was an upgrade.
Why it comes out that way
Souvlaki was a price-controlled item in those years. Its price was not set by the market but by an administrative order tied to the wholesale price of frozen beef.
The household basket in the same years was being thrown around by fresh produce: Food ran at 27.4% in 1980 against 24.7%, and 30.1% in 1981 against 24.5% for the headline index. A pinned price has no reason to follow that.
What keeping it would have cost
We measured that too before deciding: twenty years would move, the largest by 29 index points, and the peak of the series would shift from 1983 to 1988.
The headline of the site would change, on worse data.
What we kept
The Food rates stayed in the code, documented as measured and rejected, with a test that stops the conclusion from being quietly reversed.
And the same work gave us something we were not looking for: the headline CPI we already use was verified against the source, to the decimal, for twenty years.
See the same numbers live in the interactive dashboard.
The figures in this article are not written by hand: they are computed from the data series and change when it does. Prices and wages up to August 2026. Dataset version 2026.09-grammes, last updated 2026-08-31.