Season-o-meter
See how TV series are rated season by season, and film series movie by movie:
where they peaked and where they went downhill.
Live at https://seasonometer.com/, rebuilt every night
from the latest data.
Ratings come from IMDb's non-commercial datasets.
A season's rating is the average of its episode ratings. Episodes with fewer
than five votes are left out, and only series with at least 5,000 votes and two
rated seasons, or a single rated one, are included. A series with one season
has no chart: its page shows its rating and says there's nothing to compare.
IMDb doesn't say which movies belong together, so film series come from
Wikidata: everything that is a film series, film
trilogy or other kind of film series there. A film series is included if it has
at least two movies with 1,000 votes each and 20,000 votes in total.
Posters come from TMDB. This website uses TMDB
and the TMDB APIs but is not endorsed, certified, or otherwise approved by
TMDB.
Posters are shown in lists, in Swipe and at the top of each title's page.
Opening a title from a list or Swipe grows its poster into the one on its
page, in browsers with view transitions
(components/PosterTransition.tsx).
Each title's page says in a few sentences how it's rated, under the chart
(lib/summaries.ts), and ends
with six similar titles (lib/related.ts). Those links are
picked so that every title can be reached from the front page, which is how
search engines find the many pages that no list links to.
Shared links show a preview image: for a series or film series, its verdict
and a bar per season or movie, and for every other page, the site's own.
They're drawn at build time, one og.png next to each title's page
(lib/ogImage.tsx).
People
The 50 most popular directors, writers and actors each have a page with how
their movies are rated, from their first to their latest. There's no link to
them in the header. Instead, each movie on a film series' page links to the
people who made it, up to six, and each person's page to the people they
worked with most. /people/ lists them all. People can't be rated, and
neither can the movies on their pages.
They're the ones whose movies have the most votes on IMDb, with at least
five movies, counting movies with at least 10,000 votes. Writers count for
the screenplay or story, not for a book or characters a movie is based on,
or Stan Lee would be credited with every Marvel movie.
An actor's page lists every movie they're in, but cameos, billed after the
first 20 on TMDB or uncredited, are faded and left out of the ratings, the
trend and their verdict. A switch adds them to the chart. That way the
movie someone knows them from is there, without a few seconds in Hot Fuzz
counting as one of Cate Blanchett's movies. IMDb's datasets
list only about ten people per movie, in no reliable order: Cate Blanchett
isn't in two of the three Lord of the Rings movies there. So the 100 actors
with the most votes on movies they're among the first four billed in on
IMDb are looked up on TMDB, and the 50 with the most votes there, cameos
left out, are included. Without a TMDB key, it's the 50 from IMDb
(scripts/build-data.mts).
How it works
The site is fully static. There are no API calls when someone visits it,
though posters are loaded from TMDB's image server.
npm run data (scripts/build-data.mts) downloads
the IMDb datasets (~1.5 GB, cached in .imdb/ for 20 hours), queries
Wikidata for film series, looks up posters on TMDB, and writes per-season
ratings for every included series, per-movie ratings for every
included film series, and the movies of the most popular people to
data/shows.json. Posters need a TMDB API key or
read access token in TMDB_API_KEY, in .env locally and as a
repository secret for the deploy workflow. Without it the data has no
posters, and the site draws its own from the title and its ratings.npm run build reads that file and exports every page to out/:
a front page with sections for series, one for movies (/movies/) and
one for people (/people/), a page per series, film series and person,
a list page per section, a search
index for client-side search, and a sitemap so search engines find every
title's page, most of which no list links to.- GitHub Actions does both every night and
deploys to GitHub Pages at https://seasonometer.com/. The repository is
public, so Actions minutes are free. Pushes and pull requests reuse the
data from the latest nightly run instead of fetching it again; a manual
run of the workflow fetches it afresh.
The front-page sections ("Slow starters", "Collapsed like a house of cards",
…) and the verdict for each series are defined in
lib/analysis.ts, and the ones for film series in
lib/franchiseAnalysis.ts. Most verdicts have a
few wordings, listed there and in lib/verdicts.ts, and
each title always gets the same one, picked by its id. A picker at the top of
the series front page and list pages lets visitors require at least two, three
or four rated seasons in every section (four by default), except "One season
wonders", which is for series that ended after one. Every choice is rendered
at build time, and the visitor's choice is remembered in the browser.
This is the beginning of the README. Read the rest at the original. Read the full README ↗
