The system starts out with a database of singles which have been in the Top 40 of the official chart and its predecessors any time from the first UK chart in 1952 up to this time last year. The official charts currently go down as far as number 100, but I've stuck with the Top 40, partly to avoid too many obscure tracks creeping into the playlists and partly because this isn't intended as a full replica of the official chart database. Most chart music radio focusses on the Top 40 rather than tracks below it, anyway.
Once a day, the system looks for every track which was either in the Top 40 on that day in the past, or were recent new releases at that point in time in past years - that is, the sort of songs which would have been on contemporary music playlists at the time. This typically results in a list of around 3,000 songs.
It then assigns every track a weighting based on various factors. For example, songs get uprated for being new releases, new entries, big risers or number ones, and downrated for being fallers or for having lingered a long time in the chart. It also adds a small amount of random jitter so successive days aren't identical, even though they'll be based on very similar dates.
The algorithm also tries to exclude songs which wouldn't have had airplay at all, such as novelty songs, kids' songs and non-music tracks that somehow managed to chart. This isn't entirely reliable, as it relies on external data that isn't available from the Spotify API, so you may still hear something that doesn't really belong.
Once the tracks are weighted, it picks the top 100 highest rated in each category, and then randomly shuffles them to produce a playlist. A hundred songs is approximately six or seven hours listening, so if you feel like it you can listen to this most of the day and you should only hear each song once. I would call that a no-repeat guarantee, if it weren't for the fact that it's possible (though unlikely) that two different releases of the same song may show up in the same playlist.
Another drawback of the automated system is that there's no easy way to avoid tracks with explicit or offensive words. Radio stations typically play a radio edit of anything with swearing or sexual language in it, but the chart and new release data are based on the version as sold, which is the uncensored version. Again, this is something which it may be possible to address with a little more tweaking. But, at the moment, there will be NSFW songs in the playlists at times. You should be aware of that if you use the playlists in a family or work setting.
Something that is excluded, though, is certain artists who have been convicted of certain, severe criminal offences. This isn't necessarily consistent and there may be some that ought not to be there who are there. But, essentially, if they're the people that won't get played on a Friday night Top of the Pops rerun on BBC4, then they won't be here either.
A couple of additional disclaimers. Firstly, there are some charting singles which don't seem to be on Spotify - or, at least, an automated search can't find on Spotify. Where possible, the software uses the ISRC code* to look up songs, but this isn't universally available and where it isn't, the system falls back on an artist/title search. Which sometimes fails, either because the track really isn't there at all, or because the title or artist name on our database are different to those at Spotify. So there may be some tracks which ought to be here that aren't.
Secondly, along the same lines, where the system does have to use a title/artist lookup, it's possible that it may get the wrong song. There's no easy way to avoid this, although a random manual check suggests that it's rare enough to be tolerable.
The daily playlist commentary, found on each playlist page, is AI generated. At the moment, that's using the GPT-OSS-120b model, but I may experiment with other models as and when newer versions of the most popular become available. You didn't think I had time to sit and write a playlist commentary myself, five times over, every day, did you?
* Yes, I'm aware that's an example of RAS Syndrome.