According to Netflix: 'This score is unique to you, and indicates how likely we think you are to like that title.'

It is based on such factors as your streaming history, ratings you have made and the ratings of others with similar tastes.

How could the algorithm be working if I hadn't yet performed a single activity?

Netflix's ratings are based on a binary thumbs-up/thumbs-down system. You can either like or dislike a title.

I decided to give a thumbs-up to numerous titles in one area that I generally enjoy: British stand-up comedy.

Nothing changed on my homepage, so I played and skipped to the end of around 20 videos.

I began to see some recommendations. At first, they made sense...

Then, less so.

Because I 'liked' Simon Amstell (a genuine favourite), I was recommended the true-crime documentary Making a Murderer, with a 98% match. By its own definition, this means Netflix thinks I am 98% likely to like it.

It seemed like a strange connection, but then, according to Business Insider, Netflix has repeatedly claimed in the past that 'there's a difference between what you rate highly and what you actually like watching'. In this case, I concede, Netflix was correct. I have watched and enjoyed Making a Murderer.

But, then, because I 'liked' Stewart Lee (another favourite), I was recommended, with a 74% match, Pokémon the Series: XYZ, a title I doubt I would enjoy. (But, hey, maybe Netflix's algorithm knows me better than I know myself.)

At this point I had given a thumbs-up to or watched only British stand-up titles and was starting the doubt the algorithm completely.

I turned to the internet and discovered I wasn't alone. This fascinating video by theinternetftw shows how Netflix's current percentage match system fails to map to its old star rating system and, in fact, seems to 'inflate a middling or very bad score and deflate a [good score]'.

theinternetftw also discovered that there seems to be no rating lower than 55%.

I wondered if Netflix was trying to promote something in particular.

The obvious next step was to thumbs-down Netflix Original titles (those produced by Netflix itself) and to identify whether they still appeared on my homepage or were recommended; and, if so, what their percentage match scores were (presumably no higher than 55%).

I did this for 75 Netflix Original titles.

As a result, Netflix 'greyed out' the titles I rejected. The next, day, they weren't showing on my homepage at all. So far, so good.

But...

...whereas before only a handful of titles overall (whether Netflix productions or not) had been assigned percentage matches, now a great number of the remaining Netflix Original titles showed percentage matches.

And, while I didn't look at them all, the ones I did check had very high scores.

My activity to this point had been only (1) to thumbs-up and watch British stand-up titles and (2) to thumbs-down Netflix-produced shows.

After my second round of activity, Netflix began to assign ratings to many more titles across all genres. Counterintuitively, it seemed that my 'disliking' activity told the algorithm much more than my 'liking' and watching activity about what I was 'likely to like'. (Or could there have been a time lag?)

And while all the titles the system had now assigned ratings to scored very highly, it appeared that the Netflix Original titles had the highest ratings of all.

Was the binary system so broad and my activity so limited that the algorithm simply couldn't tell me much?

Or was there a strong commercial driver behind it?

My 'experiment' was far from perfect, but I would say it's at best unclear how my activity could have resulted in such high-scoring recommendations that were seemingly unrelated to the titles I had 'liked'.

The only thing I can say with a degree of certainty is that the algorithm does not consider 'Netflix Original' to be a category that can be 'disliked'. But perhaps this is fair, because it has hundreds of titles across different genres.