Formula:
score1 = ncdf(zscore(log1p(replies) - log1p(words)))
score2 = ncdf(zscore(log1p(likes) - log1p(views)))
score = max(score1, score2)
Uses a self-calibrating moving-window algorithm with N=1500 for each zscore, learns the mean and variance to ensures a [0, 100] score spread.
Also remembers the last 300 threadIds (~10 pages) to prevent recounting them repeatedly as you browse/reload.

Keeps a separate model per subforum, per website. Instantly calibrates itself to the current page on a new subforum. Browse a couple extra pages to fully calibrate it. Only works on threadmarked content, and may return nonsense scores for non-writing posts where the wordcount is meaningless.
Note: fallback to a less generous score score1
on "/search/" pages (likes unavailable).
Options:
-at the bottom, choose [unsorted|autosort]
-the [show seen|hide seen] is for my other script "Highlight visited fanfics".
-in the code, set INDICATOR2=true
to display both score1,score2 instead of the max score.
-in the code, set CORNER_TOP=false
to put indicator in bottom corner (MOBILE ONLY).
-in the code, set CORNER_INDICATOR=false
for text only scores (desktop and forum only).
code changes may be lost unless you disable updates.