A real readability score would get weightings by getting real humans to rank the difficulty of texts and then assigning weights that lead to a total score that has the same rank ordering that a human set down. I'm too lazy to do the math to do that.

Note on poetry and songs, dialogs, text with titles, etc.. Poetry and songs get unrealistically high scores because they don't split on tidy sentence boundaries. Anything with fragments, such as titles, names in a dialog will get unnaturally low scores.

Combined score is the sum of each score divided by group average, summed then divided by 5. SumOfMetrics(metric(i)/average(group))/CountOfMetrics. If an article had scores fore each metric that happed to be the same as the group average, then the composite score would be 1. So any composite score less than 1 is easy and any thing above 1 is hard and the sort should be rougly correct.

It is on my to do list to let you choose you own weights and to separate poetry from prose.

Note on copyright. All items in corpus are taken from public domain or CC sources-- for one, everything on the toki pona forum is CC. If your writings are missing, please contact me and give me permission to post your works in tp, or put that sort of notice on your website. Some items may be translations of copyrighted works, but this may be fair use, since toki pona warps the original so much that it is like saying "Whale beats crazy man" is a derivate work of Moby Dick.