Which counter designs actually get drawn

This service offers 461 counter designs. On 25 August 2026 it drew 24,285 counter images, 22,985 of which named a design. Thirty-nine distinct designs appeared in that day's traffic.

Twenty-five of those thirty-nine are in the gallery. The other fourteen are not offered there — and they account for 18,509 of the 22,985 images, or 80.5 %. All fourteen render, and every counter using one of them keeps working; what they lack is a place in the list a new counter is picked from.

The top of the list

design	images that day	bytes	in the gallery
30	8,645	3,167	no
17	4,476	43	no
26	4,048	3,731	no
48	3,024	17,798	yes
25	1,041	1,823	no

Four of the top five are not in the gallery a new counter is created from. Together they are 74.7 % of every counter image this service draws. They are addressed by design number, which the existing embed codes already carry, so the traffic arrives regardless.

Of the 461 designs that are offered, 25 were fetched at all that day. That is 5.4 %.

The 43-byte one

Design 17 looked like a bug at first: 43 bytes, image/gif, the exact size of a 1×1 transparent placeholder. The second most-used design in the whole service, apparently returning nothing.

It returns nothing on purpose. Its source is four lines long and includes img/1px.gif — it is the invisible counter, for people who want the statistics without a badge on the page. Reading the file settled in ten seconds what the byte count had made look alarming.

That check was worth doing precisely because the same question had come up here before, with a different file, and seventy-five minutes of fetching had been the long way to answer it. A surprising measurement is a reason to open the source, not a finding.

Why the gallery does not know about them

The gallery is built from a list of design numbers, and these older designs were never added to it. They date from 2012, they still render, and nothing ever connected "this design exists" to "this design is offered". Two of them even have their own dedicated cache files, so at some point they were considered important enough for special handling. The list can be extended; what was missing was the measurement that says which entries to add.

There is also a weight story in that table. The most-used design costs 3,167 bytes. The highest-placed design that people can actually choose costs 17,798 — 5.6 times as much for a picture that does the same job.

What to take from it

Design popularity here had always been read from a column on the counter row. That column records what was picked, once, possibly years ago, possibly never changed. It does not record what gets drawn. The log does, and it says something different: four fifths of the daily drawing work goes into pictures that are not in the gallery. That is a list to extend, and there is now a number to extend it by.