User-created datasets using ordinal scales (such as media ratings) have tendencies to drift or 'clump' towards the extremes and fail to be informative as possible, falling prey to ceiling effects and making it difficult to distinguish between the mediocre and truly excellent. This can be counteracted by rerating the dataset to create a uniform (and hence, informative) distribution of ratings, but such manual rerating is difficult.
So we'd like a command-line tool which consumes a list of pairs of media & ratings, then queries the user repeatedly with pairs of media to get the user's rating of which one is better, somehow modeling underlying scores while allowing for the user to be wrong in multiple comparisons and ideally picking whatever is the 'most informative' next pair to ask about to converge on accurate rankings as quickly as possible
How to get rankings: if there are 1000 media, it's impossible for me to explicitly rank a book '#952', or '#501'. Nobody has that firm a grip. Perhaps it would be better for it to ask me to compare pairs of media? Comparison is much more natural, less fatiguing, and helps my judgment by reminding me of what other media there are and how I think of them—when a terrible movie gets mentioned in the same breath as a great movie, it reminds you why one was terrible and the other great.
On a rating website, we are not interested in making fine distinctions among mediocre or trash. We are looking for interesting new candidates to consider, we are looking for the best. A skew maximizes the provided information to the reader in the region of interest of likely recommendations. So our distribution ought to throw most of our ratings into an uninformative 'meh' bucket, and spend more time on the right-tail extreme: do we think a given work an all-time masterpiece, or exceptional, or merely good?
however, are works of art only the sum of their parts? I would say no: they are something more like the multiple of their parts, and multiplying out many random variables tends to give something like a log-normal distribution instead. If art were really normally distributed, why do we so often watch an anime and say, "that had really great art and music and voice-acting but… the plot was so stupid I couldn't enjoy it at all and I have to give it a low rating and will never watch it again"?